They counted the number of mathematics journals that had been published that year
From a sample of jouurnals they estimated the average number of articles.
From a sample of articles they estimated how many new theorems per article.
By multiplying these numbers together they came up with an average ( ball park ) number of new theorems that year.